Multivariate fragility models for earthquake engineering

This paper employs a logistic regression technique to develop multivariate damage models. The models are intended for performance assessments that require the probability that structural components are in one of several damage states. As such, the developments represent an extension of the univariate fragility functions that are omnipresent in contemporary performance-based earthquake engineering. The multivariate logistic regression models that are put forward here eliminate several of the limitations of univariate fragility functions. Furthermore, the new models are readily substituted for existing fragility functions without any modifications to the existing performance-based analysis methodologies. To demonstrate the proposed modeling approach, a large number of tests of reinforced concrete shear walls are employed to develop multivariate damage models. It is observed that the drift ratio and aspect ratio of concrete shear walls are among the parameters that are most influential on the damage probabilities.